PERSONALIZING DIGITAL LIBRARIES FOR EDUCATION

Floriana Esposito, Oriana Licchelli, Pasquale Lops, Giovanni Semeraro

Dipartimento di Informatica, Università di Bari

Via E. Orabona, 4 – I70126, Bari

Keywords: Digital Library, e-Learning.

Abstract: E-Learning Systems enable students to work with electronic teaching materials, to join online courses, to

pass tests, and to communicate with other students or instructors. An important requirement of this systems

is the integration of external knowledge management resources into them. The digital libraries are helpful to

this purpose because materials of many digital libraries are valuable for learning. The availability of

teaching materials provided by an E-Learning Systems can be enlarged by reverting to materials existing in

several digital libraries. In this case, it is necessary to find the right document source and to supply the

suitable documents basing on the student requirements, also when the student model of the e-learning

system is still not available. In this paper, we have focused our attention on the use of user profiles,

generated by a personalization system (the Profile Extractor), to improve searching among digital libraries

or other generic information sources.

1 INTRODUCTION

Adaptive personalized e-learning systems could

accelerate the learning process by revealing the

strengths and weaknesses of each student. They

could dynamically plan lessons and personalize the

comunication technique and the didactic strategy. In

this systems, the Learning Management Systems

(LMS) consist of several components representing

different services to be used within a learning

environment. They enable students to work with

electronic teaching materials, to join online courses,

to pass tests, and to communicate with other students

or instructors. The availability of teaching materials

provided by a modern Learning Management

System (LMS) could be enlarged by reverting to

existing materials with the integration of external

knowledge management resources into LMS

(Rosenberg, 2001). It seems obvious that digital

libraries (DL) are predestinated for this purpose

because materials of many DL are valuable for

learning. For example, the contents of the DL of the

Association for Computing Machinery (ACM,

http://www.acm.org/dl) and the IEEE Computer

Society (http://www.computer.org/publications/dlib)

can be used for higher education and scientific

research. In this case, it is necessary to find the right

document source and to supply the suitable

documents basing on the student requirements, also

when the student model of the e-learning system is

still not available. Then the personalization of the

digital libraries access becames an important feature,

in fact, the personalization (Resnick and Varian,

1997; Riecken, 2000) involves techniques and

mechanisms to reduce the information overload and

facilitate the delivery of information that has been

customised for the preferences of individual users.

Machine Learning techniques have a significant role

to play in the personalized services of the Digital

Library. For example, many Machine Learning

techniques are well suited for transforming user-

activity data into useful preference rules as part of a

user profile (Jennings and Higuchi, 1993). Similarly,

case-based reasoning techniques can be used to

implement flexible similarity-based content retrieval

strategies (Balabanovic, 1997, Hammond et al.,

1996), and recently automated collaborative filtering

techniques have been used to transform raw user

preference data into sophisticated content filters

(Billsus and Pazzani, 1998; Konstan et al., 1997;

Perkowitz and Etzioni, 1997).

In this paper, we have focused our attention on a

generic LMS and the way to improve searching

among digital libraries or other generic information

sources, using the user profiles generated by our

personalization system (Profile Extractor).

The Profile Extractor system, based on Machine

Learning techniques, processes the logs of the user

279

Esposito F., Licchelli O., Lops P. and Semeraro G. (2004).

PERSONALIZING DIGITAL LIBRARIES FOR EDUCATION.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 279-284

 SciTePress

accessing the Digital Library and automatically

builds the user model.

2 PERSONALIZED E-LEARNING

SYSTEMS

In (Lauzon and Moore, 1989) Computer-Assisted

Learning is defined as “the delivery of educational

materials using computers as the main medium of

communication between instructor and student.” In

the decade that followed, several standards emerged

that emphasized the machine delivery of content

without an instructor. Thus, most contemporary

computer-based instruction emphasizes learner-

content interaction or learner-learner interaction.

In Intelligent Tutoring System area (Woolf,

1992), the module devoted to build the student

model is one of the major components of the

teaching system: namely, the student model module,

the pedagogical module, the domain knowledge

module, and the communication module. The

student model stores information that is specific to

each individual learner: it is “how” and “what” the

student learns or her/his errors, and the student

model plays a main role in planning the instructional

path. The pedagogical module provides a model of

the teaching process, using the student model in

order to decide the instruction method that reflects

the differing needs of each student. The domain

knowledge module contains information the tutor is

teaching, and the communication module creates the

interactions with the learner using the information

contained in the student model in order to render the

communication more effective. The information

collected during the interaction can be used to

modify the student model.

The use of student models to individualise the

interaction in hypermedia and on-line instruction

systems has been described by several authors (Bull,

1995; Bull and Smith, 1997; Smith and Jagodzinski,

1995), but the application of the techniques

suggested by ITS technology to generate the

effective presentation of instructional material has

had little practical success. According to Hartley

(Hartley, 1998), the main cause is the lack of

dialogue between researchers of the different areas,

while others think that the intrinsic complexity of

student models is at the root of the problem of their

application to configure learning for the individual

(Cummings, 1998; Ohlsson, 1993; Self, 1990).

The range of student modelling approaches

available is surveyed by Ragnemalm (Ragnemalm,

1996), who

distinguishes between models that

contain student’s actual domain knowledge and

those that contain student characteristics.

In the 1996 Vassileva (Vassileva, 1996)

describes a student model as an example of a general

user model, but where a representation of student

knowledge, held by the system, is compared with a

representation of the domain and a representation of

an expert or desired state. The aim of such systems

is to compare the student, domain and expert models

and to attempt to configure presentation of

information based in some way upon differences

between them, in order to permit to the student to

reach a desirable knowledge level (educational

goal).

In the 1996, Brusilovsky (Brusilovsky, 1996)

faces the problem to make the adaptive hypermedia

systems and states that is necessary to use some

features as goals, knowledge, background,

experience and preferences.

At present, the main engines of an e-Learning

System are the Learning Management System

(LMS) and the Learning Content Management

System (LCMS).

A Learning Management System is intended to be a

content-neutral platform that helps a learner to tailor

the delivery of instruction to individual needs. These

systems are concerned with the delivery of pre-

packaged content. They use the information in the

learner profile and learner history to select materials

from the content database. In these systems, each

instructional module and test in the content database

is tied to one or more learning objectives, material

selection and sequencing is under the control of the

instructional designers who created the original

learning modules. Most of these systems will present

a pre-test, choose material according to the test

results, present the material, then administer post-

tests and store those results as well.

The LMS integrates all aspects for managing on-

line teaching activities. From an end-user point of

view, a LMS provides an effective way to keep track

of individual skills and competencies, and provides a

means of easily locating and registering for relevant

learning activities to further improve the learner’s

skill levels.

The LCMS (Maish Nichani) offers services that

allow for the content management, creation, delivery

and reuse. Content is typically maintained in

learning objects, each of which satisfies one or more

well-defined learning objectives. A LCMS may

locate and deliver a learning object to the end-user

as an individual unit to satisfy a specific need or

may deliver the learning object as part of a larger

course, curriculum, or learning activity defined in a

LMS. Then the LMS and LCMS both monitor the

delivery of content but at different levels of

granularity. A LMS concentrates on course-level

tracking, on particularly completion status and

rolled-up scores. In contrast, an LCMS employs

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

280

detailed tracking at the learning-object level, not

only to trace user performance and interactions at a

finer granularity, but also to provide the metrics that

help authors analyze the learning object’s clarity,

relevance, and effectiveness.

Users play a central role in both LMS and

LCMS. A typical LMS maintains all information of

each user such as organizational affiliations, job

role, preferences, competencies, skill levels,

participation in past learning activities, and so forth.

Adding to this information the capability of

monitoring and planning the educational process in

order to attain learning objectives means to have an

effective student model, with the same philosophy of

a traditional computer assisted learning system.

In an e-learning system, users typically use the

LMS to manage their current competency status, to

analyze their skill gaps, and to register for learning

activities that will help them reduce their skill gaps

against an aspired career path. An LCMS focuses on

delivering a personalized experience to the user

providing just enough content to address the

person’s individual needs, when she/he needs it. An

LCMS may also enhance this experience by

customizing the content basing on a user’s profile or

by offering rich collaborative and knowledge-

exchange capabilities about the content. The key

difference is that the LCMS takes advantage of all

the information available about the user to offer a

personalized experience when delivering a learning

object, while a LMS typically maintains the user

information and makes it available to the LCMS to

deliver the personalized experience.

The LCMS can use this information also to

deliver a customized track of the learning object to

the user automatically; it can also analyze trends by

correlating the user properties from LMS and can

use it to prescribe an appropriate track to future

users, based on their profiles. The LCMS may

behave as an intelligent system that learns, based on

real data, what worked for whom and then uses this

information to help future users.

During the learning session many references

could be cited and, if possible, reproduced. The

availability of teaching materials provided by an e-

learning system can be enlarged by reverting to

materials existing in several digital libraries. In this

case, it is necessary to find the right document

source and to supply the suitable documents basing

on the student requirements, also when the student

model of the e-learning system is still not available.

The accessibility to the digital library material must

be guaranteed and a simple student model which

discovers the preferences, needs and interests of

users accessing the Digital Library should be

sufficient. The preferred searched documents supply

an interesting information for the student model too

since suggest the user main interest to deepen

specific subjects.

3 USER MODELLING FOR

ACCESSING DOCUMENT

EXTERNAL SOURCES (DL)

In our laboratory, a system has been developed to

generate user profiles automatically: the Profile

Extractor personalization system which employs

supervised learning techniques to automatically

discover user/student model through the analysis of

past user interaction with the “web” system.

The Profile Extractor is able to analyze data

gathered from sources such as data warehouse or

interactions (between student and LMS) in order to

infer rules describing the user/student behavior.

Figure 1: The learning process

XML-compliant

Learning

System

Rule

Sets

Training

Instances

Unclassified

Instances

Interaction

Data

Personal

Data

Figure 1: The learning process

PERSONALIZING DIGITAL LIBRARIES FOR EDUCATION

281

Rules are exploited to build profiles containing

preferences such as the material categories the

user/student is interested into.

Some preliminary work is needed to establish a

formal description of the features and attributes that

are needed to accomplish the given task; we can use

the results to define the representation language of

the entire learning framework. From our point of

view, the problem of learning user’s preferences can

be cast to the problem of inducing general concepts

from examples labelled as members (or non-

members) of the concepts. In this context, given a

finite set of categories of interest C = {c

, c

, …c

the task consists in learning the target concept T

“users interested in the category c

”. In the training

phase, each user represents a positive example of

users interested in the categories he or she likes and

a negative example of users interested in the

categories he or she dislikes. We chose an

operational description of the target concept T

using a collection of rules that match against the

features describing a user in order to decide if he or

she is a member of T

As depicted in Figure 1, the data about users (an

XML file containing personal and interaction data of

the user) are arranged into a set of unclassified

instances (each instance represents a user). The

subset of the instances chosen to train the learning

system has to be labeled by a domain expert, that

classifies each instance as member or non-member

of each category. The training instances are

processed by the Profile Extractor, which induces a

classification rule set for each category of interest.

More precisely, the architecture of the PE is made

up of several sub-modules:

− XML I/O Wrapper, which is the layer responsible

for the extraction of data required for the learning

process.

− Rules Manager, which is implemented through

one of the WEKA (Frank and Witten, 1998)

classifiers. The learning algorithm adopted in the

rule induction process is PART (Witten and

Frank, 1999), which produces rules from pruned

partial decision trees.

− Profile Manager, which classifies each user on

the ground of the users’ transactions and the set

of rules induced by the Rules Manager. The

classifications, together with the interaction

details of users, are gathered to form a user

profile.

Extensive experimentation of the system proposed

for the automatic extraction of the user profile has

been carried out in the digital libraries. In particular,

the system was tested on the frame of COVAX

Digital Library.

The purpose of COVAX (Contemporary Culture

Virtual Archives in XML) project (Bordoni, 2002)

was to analyse and draw up the technical solutions

required to provide access through the Internet to

homogeneously-encoded document descriptions of

archive, library and museum collections based in the

application of XML. The project demonstrated its

feasibility through a prototype containing a

meaningful sample of all the different types of

documents to build a global system for search and

retrieval.

In the experiment we considered the four main

collections of the COVAX Digital Library

(Bibliographic, Museum, Electronic Text, Archive).

For each of the 4 classes, the system was trained to

infer proper classification rules, on the basis of an

instances set representing different digital library

users given by an expert acting as a trainer of the

system. Figure 2 shows a classification rule set

example, that is generated for the “Bibliographic

Collection” on the ground of logs containing

interaction and user features; those rule sets are

expressed as disjunctions of conditions.

Figure 2: An example of classification rules for the class “Bibliographic Collection”

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

282

Figure 3: An example of a user profile

ure 3: An exam

le of a user

rofile

Using these rule sets, the classifier (Profile

Manager) predicts, for each category, whether the

user is interested or not. All these classifications,

together with the interaction details, are gathered to

form a user profile (see Figure 3 as user profile

example).

After the training phase, once a user accesses the

COVAX Digital Library, his/her dialogue history

file is generated or updated by the system. The file is

then exploited to produce a new example that the

Profile Extractor classifies on the grounds of the

rules inferred. In this way, the system is capable of

tracking user behaviour evolution and, consequently,

the profiles are updated across multiple interactions.

The concepts underlying information retrieval

were conceived long before computers and

information systems were employed to store library

materials. In the digital library domain there is a

variety of information retrieval techniques, including

metadata searching, full-text document searching,

and content searching for several data types.

The success of information retrieval can be

measured in terms of the percentage of relevant and

extraneous information retrieved, but it is difficult to

identify quantitatively the effectiveness of the

retrieval process because only an individual user can

determine what is truly useful. There are different

techniques to improve the retrieval effectiveness by

means of the extraction of additional metadata, and

recently also by means of the creation and

maintenance of user profiles.

A possible way to improve searching in COVAX

DL or in another Digital Library is to use the

information stored in a user’s profile in order to

refine the original query issued by the user. For

example, the preferred category can be enclosed in

the query submitted to COVAX search engine for

more precise result identification.

Another possible way to improve the retrieval

process (without modifying the original user’s

request) is to rank the documents in the result set

according to the categories and their degree of

interest stored in the user’s profile.

The preferred searched documents supply an

interesting information for the student model too

since suggest the user/student main interest to

deepen specific subjects.

4 SUMMARY AND OUTLOOK

A key issue when developing personalized

applications is constructing accurate and

comprehensive customer profiles based on the

collected data. User modelling is crucial for

improving the interaction between systems and their

users. In the e-learning systems, the teaching

materials provided by Learning Management System

(LMS) could be enlarged by the integration into

LMS of external knowledge management resources

as the Digital Libraries. Indeed, documents and

materials of many DL are valuable for learning and

providing intelligent personalized user support in

accessing the digital libraries, in finding the right

document source and in supplying the suitable

documents basing on the student requirements is a

main topic.

We have presented the Profile Extractor, a

system based on Machine Learning techniques,

which processes the logs of the user accessing the

PERSONALIZING DIGITAL LIBRARIES FOR EDUCATION

283

Digital Library and automatically builds the user

model. The automatic generation and discovery of

the user profile allows to improve searching among

extremely large Web repositories, such as Digital

Libraries or other generic information sources, by

providing them with personal recommendations.

By the e-learning system perspective, this profile

con

ACKNOWLEDGEMENT

This work was partially supported by ENEA under

REFERENCES

Balabanovic, M. and Shoham, Y., 1997. Fab: Content-

Bil 8. Learning

Bru s and techniques in adaptive

Bo Culture

Bull, S., Brna, P, and Pain, H., 1995. Extending the scope

Bu student models to

Cummings, G., 1998. Artificial intelligence in education:

an exploration. Journal of Computer Assisted

Learning, 14(4) 252-259.

Experiences Lessons and

Har

, 1-25.

net News.

Lau

computer

ng, 9(4) 194-221.

onference on Artificial

Rag

action, 5, 93-116.

lized Views of Personalization.

Ros

egies for

oads of artificial intelligence and education.

earning Technology

Vas

odelling and User-Adapted

stitutes a first student model, based on

communication preferences more then on learning

performances, useful to create a personalized

education environment. The future step is trying to

use such information in order to plan a personalized

educational path, constantly monitoring the

educational process. The student model constructed

initially can be refined and/or reviewed on the basis

of the new inputs to the system. Once again Machine

Learning techniques could be of use in the automatic

review of student models, by offering incremental

learning methods in order to update the knowledge

learned on the basis of new observations.

grant “An innovative system for the extraction of

user profiles”. The authors would like to thank

Luciana Bordoni and Fabrizio Poggi for the dataset

used in the Profile Extractor.

Based, Collaborative recommendation.

Communications of ACM, 40 (3),66-72.

lsus, D. and Pazzani, M. J., 199

Rie

Collaborative Information Filters. In Proceedings of

the International Conference on Machine Learning,

Wisconsin, USA, 46-54.

silovsky, P., 1996. Method

hypermedia. User Modelling and User-Adapted

Interaction, 6(2-3), 87-129.

rdoni, L., 2002. COVAX: A Contemporary

Virtual Archive in XML. In Proceedings of the 6th

European Conference ECDL 2002, Rome, Italy, 661-

662.

of the student model. User Modelling and User-

Adapted Interaction, 5(10), 45-65.

ll, S. and Smith, M., 1997. A pair of

encourage collaboration. In Proceedings of the 6th

International Conference on User Modeling UM97,

Italia, 339-341.

Hammond, K. J., Burke, R. and Schmitt, K., 1996. A

Case-Based Approach to Knowledge Navigation.

Case-Based Reasoning

Future Directions, MIT Press, 125-136.

tley, J.R., 1998. Ospite Editoriale: CAL and AI - a

time for rapprochement? Journal of Computer

Assisted Learning, 14(4), 249-250.

Jennings, A. and Higuchi, H., 1993. A user model neural

network for a personal news service. User Modeling

and User-Adapted Information, 3(1)

Konstan, J. A., Miller, B. N., Maltz, D., Herlochker, J. L.,

Gordan, L. R. and Riedl, J., 1997. Grouplens:

Applying Collaborative Filtering to Use

Communications of the ACM, 40(3), 77-87.

zon, A.C. and Moore, G.A.B., 1989. A fourth

generation distance education system: Integrating

computer-assisted learning and

conferencing. The American Journal of Distance

Education, 3(1), 38-49.

ish Nichani. LCMS = LMS + CMS

http://www.elearningpost.com/features/archives/00102

2.asp

Ohlsson, S., 1993. Impact of cognitive theory on the

practice of authoring. Journal of Computer Assisted

Learni

Perkowitz, M. and Etzioni, O., 1997. Adaptive Web Sites:

An AI Challenge. In Proceedings of the 15th

International Joint C

Intelligence, Nagoya, Japan. Morgan Kaufmann, 16-

23.

nemalm, E.L., 1996. Student Diagnosis in Practice;

Bridging a Gap. User Modelling and User-Adapted

Inter

Resnick, P. and Varian, H., 1997. Recommender Systems.

Communications of the ACM, 40(3), 56-58.

cken, D., 2000. Persona

Communications of the ACM, 43(8), 27-28.

enberg, M.J., 2001.

e-Learning – Strat

Delivering Knowledge in the Digital Age. McGraw-

Hill.

Self, J.A., 1990. Bypassing the intractable problem of

student modelling. Intelligent tutoring systems: at the

crossr

Frasson, C., Gauthier, G. (eds.), Ablex Publishing,

Norwood, New Jersey, 107-123.

ith, C. and Jagodzinski, P., 1995. The implementation

of a multimedia learning environment for graduate

civil engineers. Association for L

Journal, 3(1) 29-39.

sileva, J., 1996. A task-centred approach for user

modelling in a hypermedia office documentation

system. User M

Interaction, 6(2-3) 185-223.

olf, B., 1992. AI in Education. Encyclopedia of

Artificial ntelligence. Shapiro, S. ed., John Wiley &

Sons, Inc., New York, 434-44

ICEIS 2004 - HUMAN-COMPUTER INTERACTION

284